Language & Communication

Transcription

1 Language & Communication 33 (2013) Contents lists available at SciVerse ScienceDirect Language & Communication journal homepage: Variation in handshape and orientation in British Sign Language: The case of the 1 hand configuration Jordan Fenlon a,, Adam Schembri b, Ramas Rentelis a, Kearsy Cormier a a Deafness, Cognition & Language Research Centre, University College London, UK b La Trobe University, Melbourne, Australia article info abstract Article history: Available online 30 October 2012 Keywords: Sign language Sociolinguistic variation Phonology Pointing This paper investigates phonological variation in British Sign Language (BSL) signs produced with a 1 hand configuration in citation form. Multivariate analyses of 2084 tokens reveals that handshape variation in these signs is constrained by linguistic factors (e.g., the preceding and following phonological environment, grammatical category, indexicality, lexical frequency). The only significant social factor was region. For the subset of signs where orientation was also investigated, only grammatical function was important (the surrounding phonological environment and social factors were not significant). The implications for an understanding of pointing signs in signed languages are discussed. Ó 2012 Elsevier Ltd. All rights reserved. 1. Introduction In this paper, findings from the first major study on phonological variation in British Sign Language (BSL) are presented. Using data collected as part of the BSL Corpus Project (Schembri et al., 2011), we examine sociolinguistic variation in signs with the 1 handshape (i.e., with an extended index finger), using methodology adapted from work by Bayley et al. (2002) on American Sign Language (ASL). In their study, Bayley et al. found that the most important factor influencing this phonological variation was grammatical category but that a number of other factors, including features of the immediate phonological environment and social factors (e.g., age, ethnicity, language background, and region) played a role. For the current study, we investigate if the same factors influencing 1 handshape variation in ASL are also at work in an unrelated sign language and whether other factors not considered by Bayley et al., such as lexical frequency or indexicality, might also be involved. We use a variationist approach to investigate both the social and linguistic factors that may be relevant in conditioning variation (e.g., Walker, 2010). Additionally, since we are particularly interested in the pointing signs in our data, we have further examined all pronominal and locative pointing signs for variation in orientation and possible conditioning linguistic and social factors. In the following sections of the paper, we first provide some background about sociolinguistic variation in BSL, before discussing studies of phonological variation in sign languages generally. Next, we present a summary of previous work on 1 handshape variation in ASL, before discussing the BSL Corpus Project data and our studies into handshape and orientation variation. Lastly, we discuss the implications of our work for an understanding of phonological variation in sign languages and its relationship to gesture, lexical frequency and models of the sign language lexicon. Corresponding author. Address: Deafness, Cognition & Language Research Centre, University College London, 49 Gordon Square, London WC1H 0PD, England, UK. Tel.: +44 (0) ; fax: +44 (0) addresses: j.fenlon@ucl.ac.uk (J. Fenlon), a.schembri@latrobe.edu.au (A. Schembri), hipas8@yahoo.com (R. Rentelis), k.cormier@ucl.ac.uk (K. Cormier) /$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.

2 70 J. Fenlon et al. / Language & Communication 33 (2013) TALK WORK Fig. 1. A minimal pair in BSL. 2. Background 2.1. Sociolinguistic variation and change in BSL Research conducted into sociolinguistic variation in BSL has concentrated almost exclusively on the lexical level. For example, it is well documented that there is considerable regional variation in the BSL lexicon, with traditional signs for colour terms and numerals, for example, often being specific to a region in the UK (Brien, 1992; Deuchar, 1984; Sutton-Spence and Woll, 1999; Sutton-Spence et al., 1990). More recent research has, however, revealed that the use of traditional regional signs for countries, numbers and colours, appears to be diminishing (Stamp et al., in press). Stamp et al. suggest that these linguistic changes reflect social changes in the deaf community and in the use of communications technology, with greater exposure to sign variants on the internet and television and deaf people generally being more mobile, both nationally and internationally, than in the past. To date, the largest study of sociolinguistic variation in BSL focuses on fingerspelling (i.e., the use of a two-handed manual alphabet to spell out English lexical items, a system used to enable the borrowing of words from the surrounding spoken language) (Sutton-Spence et al., 1990). Sutton-Spence et al. used a dataset of 19,450 fingerspelled items collected from 485 interviews with BSL signers on the British deaf television programme See Hear. They found age and region to be significant factors in predicting use of the two-handed manual alphabet with younger signers and signers from the south-western region of England producing the least fingerspelling. No large-scale studies have as yet focused on variation in BSL at the phonological level. The research conducted here is thus the first attempt to investigate phonological variation in BSL and its possible relationship to linguistic and social factors Sign language phonology William Stokoe (1960) was the first to propose that the manual signs in sign languages could be analysed into three main sublexical elements: handshape, location, and movement. These contrastive components may be considered analogous to the parameters of speech production, such as voicing, place and manner of articulation. BSL signs are made from the combination of a limited set of parameter values (for example, there appear to be around 35 distinctive handshapes in the language), and minimal pairs may be distinguished on the basis of differences in these parameters (Brennan et al., 1984). For example, the BSL signs TALK 1 and WORK have the same location (i.e., they are articulated on the radial side of the non-dominant hand in neutral space) and the same movement (i.e., a repeated tapping movement) but differ in handshape (i.e., TALK is produced with the 1 handshape whilst WORK is produced with the B handshape) as shown in Fig. 1 below. Although there is debate about the relationship between these three parameters and additional formational elements (such as the orientation of the hands, see Sandler and Lillo-Martin, 2006) and about the nature of minimal pairs in sign languages (Liddell and Johnson, 1989), evidence for the sublexical compositionality of signs is well-established. This includes notions of well-formedness of signs shared by native signers and documented in dictionaries (e.g., Brien, 1992 for BSL), the relevance of these parameters for sign language processing (e.g., Orfanidou et al., 2010), and stages of phonological development in sign language acquisition (e.g., Mann et al., 2010). Since the 1970s, phonologists have also begun to propose models in which the parameters of handshape, location and movement can be represented as bundles of distinctive features (e.g., Brentari, 1998; Crasborn, 2001; Sandler, 1989). Handshapes, for example, can be analysed as in terms of the behaviour 1 As is conventional in the sign linguistics literature, glosses for lexical signs are given in small caps (e.g., GIVE). For signs requiring more than one English word, a hyphen is used (e.g., NEXT-YEAR). All the lexical signs in this study were assigned an ID-gloss i.e., a unique label used to identify a specific lexeme and all its phonological and morphological variants (see Section for more information about ID glossing). All the glosses used throughout this paper represent ID glosses where possible (i.e., except when discussing other studies where this approach to identifying lexical signs was not used). Pointing signs are glossed using a prefix (PT) followed by a colon and information about its function. For example, PT:PRO1, PT:PRO2, PT:PRO3 all indicate a pointing sign that functions as a first, second, or third person singular pronoun respectively. Additional glosses used with pointing signs include PT:DET, for points functioning as determiners, and PT:LOC for locative points. If a pointing sign is identified as a plural point, a -PL suffix is appended to the gloss (e.g., PT:PRO3PL).

3 J. Fenlon et al. / Language & Communication 33 (2013) Table 1 Distribution of participants according to six social categories. Site Gender Age Language background Social class Ethnicity F M Deaf Hearing Working Middle White Other Belfast (n = 30) Birmingham (n = 30) Bristol (n = 32) Cardiff (n = 29) Glasgow (n = 30) London (n = 30) Manchester (n = 30) Total of (a) the thumb: whether it is extended away from the rest of the hand, for example, or held across the other fingers, and (b) the fingers: which of the fingers are selected (i.e., actively involved in the articulation of a sign or not) and whether they are spread apart or together, bent at specific joints or fully extended Overview of phonological variation and change in signed languages Although there have been a number of studies on phonological variation in signed languages, many of these studies are small both in number and scale when compared to studies on phonological variation in spoken languages. Furthermore, since the first study published by Battison (1974), work on phonological variation has focused almost exclusively on ASL. It is only recently that studies on phonological variation in other sign languages have begun to be undertaken, including work on Australian Sign Language (Auslan) and New Zealand Sign Language (NZSL) (Schembri et al., 2009). Together, these studies begin to provide us with a cross-linguistic perspective on phonological variation in sign languages. Previous investigations have underlined the influence of linguistic and social factors in sign language phonological variation. For example, Battison (1974) collected data on the deletion of the subordinate hand in two-handed signs (known as weak drop see Brentari, 1998) and found that this process was more common in symmetrical than asymmetrical twohanded signs. Bayley et al. (2002) describe several studies (dating back to the 1970s) examining phonological variation in ASL such as Woodward et al. s (1976) investigation into the variable use of ASL signs that have related forms produced on the face or on the hands. Drawing on questionnaire data from 45 participants, these researchers found evidence of variation due to ethnicity, with black signers much more likely to use the hand rather than face variants. Their data also suggested regional differences, with signers in New Orleans producing fewer variants on the hands than those in Atlanta. Another study by Woodward and DeSantis (1977) found ethnic variation in two-handed versus one-handed forms of some ASL signs, with white signers using significantly more of the one-handed variants of these signs. They also found that signers from the southern states of the US used more two-handed variants than non-southerners, and that older signers used more than younger signers. Later, accounts of phonological variation in signed languages emphasised the key role that the immediate phonological environment plays in this variation with respect to all three major parameters (e.g., Liddell and Johnson, 1989; Johnston, 1989). That is, phonological features of a given sign, such as handshape and location, are likely to assimilate to the features of neighbouring signs, causing variation away from the citation form (i.e., the form which is produced in isolation and listed in dictionaries). Although the importance of the immediate phonological environment for phonological variation has been given empirical weight in recent analyses using large corpora which have focussed on 1 handshape variation (e.g., Bayley et al., 2000, 2002) and variation in signs produced at the forehead location (e.g., Lucas et al., 2002; Schembri et al., 2009), these studies demonstrate that other factors in addition to the phonological environment may be at work in conditioning variation. For example, depending on the target variable, some grammatical categories of signs (e.g., pronouns or verbs) can be expected to vary more compared to others. Schembri et al. (2009), in applying the work of Lucas et al. (2002) to Auslan and NZSL data, initially found grammatical category to be the most important factor but discovered an interaction between location variation and lexical frequency. Highly frequent verbs were found to occur significantly more frequently in lowered form (i.e., at locations lower than the forehead) than all other sign types (e.g., high frequency nouns and adjectives and low frequency nouns, adjectives and verbs). This observation about the relationship between lexical frequency and phonological variation runs parallel with findings in spoken languages (Bybee, 2002; Philips, 1984). In addition to linguistic factors, the studies mentioned above have demonstrated that social factors have an important role in phonological variation in signed languages. Social factors found to be relevant include region, age, gender, language background (i.e., whether a participant was raised in a deaf signing family or not), ethnicity, and social class. These social factors and the extent to which they influence phonological variation, however, appear to vary across sign languages. For example, region and gender were found to be important factors in location variation for the Auslan and NZSL location studies. Age, however, was only a significant factor in Auslan, and language background and ethnicity were found to be significant in NZSL only. Furthermore, although the NZSL and ASL studies of location variation both found language background and ethnicity were important, their results indicated different effects in each community (e.g., native signers used significantly more citation forms in ASL while native signers tended to use more variant forms in NZSL).

4 72 J. Fenlon et al. / Language & Communication 33 (2013) Table 2 Handshape categories representing possible phonological variants of the 1 handshape. a Handshape variant: Definition: c Citation form: thumb, -fingers, -bent, +index (1) X thumb, fingers, +bent, +index (X) 6 +thumb, fingers, +bent, index (6) & +thumb, fingers, +bent, +index (&) L +thumb, fingers, bent, +index (G, L) 4 thumb, +fingers, +/ bent, +index (4, 4, H) 5 +thumb, +fingers, +/ bent, +index (B, 5, 9, 8, 7, ILY, 3, bo) o Some other handshape a The handshape images shown in this paper are from the Hamburg Notation System (Prillwitz et al., 1989). Similar observations reported in the studies above have been made for BSL but on the basis of relatively little data. Deuchar (1981) claimed that the deletion of the non-dominant hand in symmetrical two-handed signs, such as GIVE and HOSPITAL, was frequent, as also noted in ASL (Battison, 1974). Deuchar argued that this weak drop in asymmetrical two-handed signs appeared most likely in signs where the handshape on the non-dominant hand was a relatively unmarked configuration, such as the B handshape (see illustration of this handshape in Table 2). Thus, variants without the non-dominant hand

5 J. Fenlon et al. / Language & Communication 33 (2013) seemed more common in her data in signs such as RIGHT (with a non-dominant B handshape) than in FATHER (non-dominant H; see Table 2 for illustration). In a small pilot study, Deuchar also found that discourse factors underlie the frequency of this phenomenon with informal varieties producing more tokens of weak drop than formal varieties. Deuchar (1984) later suggested that this weak drop variation may also reflect language change in progress, based on Woll s (1981) claim that certain signs (e.g., AGAIN) which appear to be now primarily one-handed in modern BSL were formerly two-handed. Furthermore, Deuchar (1984) notes that progressive and regressive assimilation of handshape is possible in BSL and suggests that there may be underlying grammatical constraints at work. However, a thorough analysis of how such phonological processes work in BSL has yet to be undertaken. Despite considerable work on the theoretical aspects of phonology in the field of sign language linguistics (e.g., Brentari, 1998; Crasborn, 2001; Sandler and Lillo-Martin, 2006), there remain relatively few studies examining sociolinguistic variation at the phonological level in signed languages. A thorough cross-linguistic perspective on sociolinguistic variation and change in phonology is needed, however, so that we can build a better understanding of what kind of internal and external factors constrain phonological variation in signed languages, and how these factors compare to those found in spoken languages Variation in the 1 handshape In an influential paper on ASL phonology, Liddell and Johnson (1989:250) made a specific claim about signs with the 1 handshape. They observed that the hand configuration of the first person pronominal sign in ASL (which they referred to as ME; the form of this sign is identical to PT:PRO1 that we report here for BSL) typically assimilates to that of a contiguous predicate in the same clause. They went on to claim that the extent to which signs other than ME assimilate to the hand configuration of another sign, although not yet thoroughly investigated, appears to be considerably more limited. This claim was investigated for the first time by a large-scale study examining variation in the 1 handshape in ASL conducted by Bayley et al. (2002). They collected 5356 sign tokens from 207 signers engaging in spontaneous conversation (filmed in groups or in pairs). Their pool of participants consisted of native or near native signers (those who had learnt to sign before the age of 6) in a quota sample mixed for gender, age, language background, ethnicity and social class and recruited from seven sites around the United States: Staunton, Virginia; Frederick, Maryland; Boston, Massachusetts; Olathe, Kansas; New Orleans, Louisiana; Fremont, California; and Bellingham, Washington. An analysis of 5195 tokens, excluding the thumb-only variant, 2 revealed that signs that used the 1 handshape in citation form were more often realised with some other handshape (60% of their data). This suggested that this variation was not as limited an occurrence as previously claimed by Liddell and Johnson (1989), although the first person pronoun did indeed account for a considerable proportion of this variation. Despite this large amount of variation, almost all of the variant tokens could be grouped into just two main handshape categories: the L handshape (30%), and the 5 handshape (25%). The remaining 5% of their data was divided between two other variants: those with the 4 handshape and the X handshape (see Table 2 for illustrations of these handshapes). Each token was analysed for a number of linguistic and social factors. Results indicated that grammatical category was the most important factor conditioning variation in the 1 handshape in ASL. Specifically, nouns, adjectives, verbs, adverbs, wh-signs, and third person pronouns (singular and plural forms of all pronouns were not distinguished in the analysis) appeared significantly more frequently in citation form. Conversely, first and second person pronouns tended to appear more often in non-citation form (i.e., using some handshape other than the 1 handshape). This finding led Bayley et al. to conclude that there is a systematic correspondence between indexicality and the degree of variation observed in handshape. Specifically, Bayley et al. argued that first person pronouns, the category that was found to vary most over others, is also the category considered to be the most indexic (both singular and plural forms involve points towards the chest of the signer). This is followed by second person pronouns, the next category found to vary the most, which they analysed as less indexic than first person pronouns but more so than third person pronouns (typically, the singular and plural forms both involve a point towards the signer s conversational partner(s)). In contrast, third person pronouns, signified by a point that indicates a non-addressed referent or group of referents (whether physically present or not), Bayley et al. propose, are the least indexic of the three and tend to favour citation form in the ASL data. Finally, Bayley et al. note that in non-indexical grammatical and content signs, the handshape has more semantic relevance and therefore is likely to retain its citation form. Additionally, whilst all categories of non-pronominal signs were observed to vary in handshape, they were less likely to show the same degree of handshape assimilation as pronouns (i.e., there was assimilation in fewer of the distinctive features in the handshape of signs in these other categories) since doing so may have a relatively greater effect on its perceived meaning (i.e., the sign may be interpreted as a different sign by the conversational partner). Bayley et al. (2002) also found features of the immediate phonological environment to have a significant effect on 1 handshape variation. In many tokens, handshape features of the target sign appeared to show assimilation to features of the signs before and after the target sign. That is, if the handshape before was articulated with an extended thumb (i.e., open and extended to the side of the hand), it was likely to observe a variant with an extended thumb in the target sign. Conversely, if the handshape before was articulated with an opposed thumb (i.e., a thumb held across the fingers), it was more likely to 2 Bayley et al. (2002) observed that the thumb-only variant of a pronominal pointing sign could be influenced by the physical location of its (second or third person) referent. As they filmed people in groups, a point (using the 1 handshape) to a referent situated to the signer s immediate ipsilateral side (to the right for right-handed signers) may be physically uncomfortable and the thumb-only variant might be produced instead. 161 tokens were excluded from their analysis for this reason (5356 tokens originally).

6 74 J. Fenlon et al. / Language & Communication 33 (2013) observe an opposed thumb variant in the target sign. Bayley et al. found a clear relationship between the number of features the citation form of the target sign shares with adjacent signs and the likelihood that it will retain its citation form. An analysis of this finding alongside grammatical category suggests that the two categories are working in tandem. That is, when more features are shared with the target sign as a noun it is more likely to keep its citation form than when it is a first person pronoun. In addition to linguistic factors, Bayley et al. found a range of social factors to be significant factors conditioning variation in the 1 handshape in ASL: region, age, language background, social class and ethnicity. The remaining social factor in their analysis, gender, was not found to be significant. Results by region tended to fall into two groups. Signers from California, Kansas/Missouri, Louisiana and Massachusetts all used significantly more citation forms, whilst signers from Maryland, Virginia and Washington used fewer citation forms. Within this regional grouping, further interaction with the remaining significant social factors was observed. Younger signers from California, Kansas/Missouri, Louisiana and Massachusetts used significantly more citation forms, for example, whereas older signers did not (although this was not a strong effect). Similarly, middle class signers in Maryland, Virginia and Washington used more citation forms than working class signers. In contrast, working class signers in California, Kansas/Missouri, Louisiana and Massachusetts produced more examples of the citation form than middle class signers. The results for language background as a significant factor were mixed. When a specific handshape variant was selected as the dependent variable (the 5 handshape variant in which all fingers and the thumb were extended), signers with deaf signing parents in California, Kansas/Missouri, Louisiana, and Massachusetts were found to use significantly fewer tokens of this variant, whilst signers with hearing parents in the same areas used more. Language background was not found to be significant in analysis restricted to Maryland, Virginia and Washington alone. Finally, ethnicity was found to be a significant factor influencing variation in the 1 handshape: black Americans used significantly more examples of the citation form compared to white Americans. However, the authors conclude that this result is a reflection of regional differences in preference for citation form rather than ethnicity itself. That is, the number of African Americans in areas that tended not to use citation forms (e.g., Maryland, Virginia and Washington) was so low that a meaningful analysis based on ethnicity could not be achieved. Although sign language phonologists had observed these patterns previously (e.g., Corina, 1990; Liddell and Johnson, 1989; Sandler, 1999), Bayley et al. (2002) was the first study to provide empirical evidence from a large naturalistic dataset for handshape assimilation effects in any signed language. Furthermore, it was one of the first to show that grammatical function appeared to be a significant factor conditioning the frequency of variant forms. The work of Bayley et al., however, did not include any investigation into the possible effect of individual lexical items on the data. For example, although the results for specific person forms for pronouns are presented, singular and plural forms of the first person pronoun are grouped together. Guy (2007) points out that research on phonological variation in spoken languages has shown that phonological variation does not affect all lexical items sharing particular phonological features in the same way (for example, final stop deletion in the English word and is much more frequent than other words with a word final stop). One of the key factors related to these lexical effects in phonological variation is lexical frequency. Bayley et al. did not investigate frequency effects in their data, despite the fact that singular and plural forms of the first person pronoun represented almost 40% of their entire dataset. Furthermore, they claim that there is a relationship between indexicality and variation in the first, second and third person forms. In order to strengthen this claim, there is a need to separate out plural and singular forms of pronominal signs, as there is some evidence that plural pronoun forms are less indexic than singular forms in both ASL and BSL (Cormier, 2007). With regard to the influence of social factors, Bayley et al. (2002) identified significantly different degrees of handshape variation in different regional groupings due to social class and age. Despite this, Bayley et al. (p. 49) stressed the near uniformity of constraint effects across regional and social groups that illustrated that ASL users constituted a single language community. They suggested that differences between some social subgroups in the patterns of 1 handshape variation in first person and non-first person pronouns (with significantly greater handshape assimilation in the first compared to non-first person pronominals) reflects the possibility that grammaticalisation of a first versus non-first person distinction is at work, and that it is unfolding at different rates in different subsets of the American deaf community. 3 An issue not considered by Bayley et al., however, is the degree to which such variation also occurs in pointing gestures, particularly between pointing gestures to the self and to others. An analysis of 388 body-directed gestures by Cooperrider (2011), collected from a set of televised interviews with 40 native speakers of American English, indicated that only 10% of self-points used a handshape with an extended index finger (with or without thumb extension). The figure in the ASL data was different, with about 48% of all first person pronouns using the 1 handshape (thumb opposed), or the related L handshape (thumb extended). Handshape variants other than a 1 or L handshape accounted for 52% of the ASL data, but almost 90% of the gesture data. Unfortunately, however, Cooperrider did not undertake studies of handshape assimilation effects in his gesture data, nor do we have comparable figures for pointing gestures to non-first person referents. Nevertheless, the grammaticalisation hypothesis needs to take into account the patterns of variation in related pointing gestures used by non-signers which are undoubtedly related to these forms in sign languages, both in form and function (Cormier et al., submitted for publication; Johnston, in press). 3 This is in contrast to Meier s (1990) claim that a (presumably already uniform and stable) grammatical distinction exists between first and non-first person in the ASL pronominal system.

7 J. Fenlon et al. / Language & Communication 33 (2013) Variation in orientation in pointing signs A large proportion of the signs with a 1 handshape that appear in sign language discourse involve pointing signs of various types (Johnston, 2012; McKee and Kennedy, 2006; Morford and MacFarlane, 2003). This appears to be especially true of the spontaneous conversational BSL data used in this study (Cormier et al., 2011). A number of researchers have suggested that variation in the orientation of pointing signs is conditioned by the grammatical function of the pointing sign. Pfau (2010) provides an overview of pointing in sign languages, showing how a number of researchers working on a range of sign languages (de Vos, 2008; Engberg-Pederson, 2003; van der Kooij et al., 2006) have proposed that pronominal or anaphoric pointing tends to occur with a palm vertical orientation, whereas locative pointing favours palm down orientation. To our knowledge, there has been no empirical investigation of these claims (certainly not with a large dataset), nor has the relationship to similar patterns in co-speech pointing gestures been explored (similar claims about the relationship between different indexical functions and orientation has been made for Neapolitan gesture, see Kendon and Versante, 2003). 3. Research questions We had two main research questions that motivated our study of sociolinguistic variation and change in the 1 hand configuration in BSL. First, we wanted to know what conditions handshape variation in signs that have the 1 handshape in citation form in BSL compared to ASL, given that the two sign languages are historically unrelated. Would the same linguistic and social factors be involved? Would additional factors, such as lexical frequency, be important? Second, we were interested in understanding orientation variation in signs with the 1 handshape in BSL: Which linguistic and/or social factors would condition orientation variation in BSL? Also, would the analysis reveal the kind of systematic difference between pronominal and locative pointing signs that has been suggested in the sign language literature for other sign languages? 4. Methodology 4.1. BSL Corpus Project The study reported in this paper draws on data collected as part of the BSL Corpus Project (Schembri et al., 2011), a largescale project that aimed to create the first online, open-access corpus of BSL. The BSL Corpus features digital video data collected from 249 deaf signers from eight urban centres around the United Kingdom. All participants were filmed while engaged in four tasks: retelling a personal experience narrative, participating in a 30-min conversation, answering questions on language attitudes and awareness, and responding to a short lexical elicitation task. The large, semi-stratified sample and the type of data collected makes the BSL Corpus dataset an ideal resource for the questions put forward in this study. For this study, we focussed on data from 211 signers across 7 cities. The distribution of participants according to several social categories is provided in Table 1. In the following sections, the BSL Corpus Project methodology in regard to these categories is described in more detail Sites Phonological variation due to region is a widely-studied phenomenon in British English (for an overview, see Trudgill (1999)). We planned to include region as a social factor in our analysis of 1 handshape variation because results from previous studies in phonological variation in other signed languages (e.g., Bayley et al., 2002; Lucas et al., 2002; Schembri et al., 2009) have suggested that region is an important factor in conditioning variation in signed languages as well. For this study, we included data from 7 cities representing some of the major urban centres in the four regions of the United Kingdom: Birmingham (in central England), Bristol (in the southwest), London (in the southeast), and Manchester (in the northwest); Belfast in Northern Ireland; Glasgow in Scotland; and Cardiff in Wales. To ensure that each participant was representative of their region, only participants who had lived or worked in that region for ten years or more were invited to take part in the BSL Corpus Project Participants A total of 211 deaf participants were included in the study. In terms of language background, 94% (n = 199) of participants included in this study reported learning to sign before the age of 7 and all the remaining individuals reported that they learnt to sign by the age of 12. Native signers (i.e., those from a deaf family exposed to BSL from birth) represented 32% (n = 68) of participants. Research has shown that the age at which a child is exposed to sign language as a first language has a considerable effect on their sign language proficiency in adulthood (for a review, see Emmorey, 2002). The participant sample was balanced for gender. Of the 211 participants, 52% (n = 109) were women. It is well documented that gender is a significant factor in phonological variation in spoken languages (e.g., Cheshire, 2002). This has also been found for signed languages (e.g., Lucas et al., 2002; Schembri et al., 2009). Age-related variation is well documented for spoken languages (e.g., Bailey, 2002) and for signed languages, including BSL (Lucas et al., 2002; Stamp et al., in press; Sutton-Spence et al., 1990). Due to variable patterns of language transmission within the deaf community (the language is sometimes learned from peers in schools for deaf children, or after school as a young

8 76 J. Fenlon et al. / Language & Communication 33 (2013) adult in other settings), clear differences can be seen between older and younger signers (for example, in the use of lexical items, as explained above, see Sutton-Spence et al., 1990; Stamp et al., in press). Recruitment to the BSL Corpus Project was designed to reflect this variation by ensuring that participant selection was spread across age groups, ranging from 16 to 94 years of age. For the study reported here, participants were grouped into two categories according to age: younger (16 50) and older (50 94). The division of participants into these age groups is partly motivated by changes in language policy in deaf education during the twentieth century (e.g., from education that emphasised the exclusive acquisition of speech and listening skills to increasing acceptance of sign language in the classroom; see Woll and Ladd (2011) for an overview). Phonological variation due to social class is well-established in spoken languages (e.g., Ash, 2002), and there is some evidence for it being a relevant social factor for phonological variation in ASL, although often in interaction with other social factors. It was found to be relevant for an understanding of 1 handshape variation in ASL, although in this case (as described above), it exhibited a complex interaction with region (Bayley et al., 2002). For the current study, we classified participants into two broad social classes based on their occupation and/or educational background. Deaf individuals with a university education and/or white collar professional occupations were categorised as middle class, whereas individuals with no university education and having traditional blue-collar factory or trade-related occupations were classified as working class. Ethnicity is an important social factor in sociolinguistic variation in many English-speaking communities (Fought, 2002), and has been shown to be relevant to studies of phonological variation in ASL (Lucas et al., 2002). The ethnic composition of the British deaf community is, however, unknown. As the overall British population was reported to be about 10% non-white (mostly of Afro-Caribbean and south Asian origin) in the 2001 Census (Office for National Statistics, 2006), we attempted to include a similar proportion of non-white participants in our project Data collection Data collected as part of the BSL Corpus Project involved working with a deaf fieldworker from each site who conducted participant recruitment (a methodology first used in deaf communities by Ceil Lucas and colleagues, see Lucas et al., 2001). All fieldworkers were deaf and all but one were native signers. All fieldworkers were also present on the day of filming and assisted in data collection together with a deaf project researcher. No hearing people were present during filming. This kept possible language contact influences to a minimum (Lucas and Valli, 1992). Participants were always filmed in pairs and, where possible, were filmed with another person within, or close to, their age group. As participants were required to hold a conversation for 30 min, they were partnered with someone familiar (i.e., a friend) so that they felt relaxed and any awkwardness could be avoided. Filming sessions took place in settings familiar to the participants, e.g., deaf centres and deaf organisations, to ensure that participants felt comfortable and that it was appropriate to use a relatively informal variety of BSL Data coding For the 1 handshape variation study, we coded 10 sign tokens each from 211 participants participating in the first, more informal half of the filming session, which included participants recounting personal narratives and/or engaging in free conversation. These sign tokens were signs that were known to have the 1 handshape in their citation form. These tokens include pronominal pointing signs (i.e., first, second, and third person pronouns), locative pointing signs (e.g., PT:LOC meaning there or here ), wh-signs (e.g., WHAT, WHO, WHY), conjunctions (e.g., BUT, AND) as well as nouns (e.g., INFORMATION, ENG- LAND, GIRL), verbs (e.g., TALK, DEBATE, SAY) and adjectives (e.g., QUICK, SLOW). All the lexical signs in this study were assigned an ID gloss, which represents emerging best practice when annotating a sign language corpus (Johnston, 2010). An ID gloss is a unique label used to identify a particular lexeme and to represent all its phonological and morphological variants in the process of lemmatisation. ID glosses do not reflect the meaning of a sign across all contexts, nor do they give any indication of a token s grammatical function. For example, the ID gloss ENGLAND is used for the sign which can mean England or English, regardless of whether the token in question is functioning as a noun or adjective or whether it refers to the country, culture, language, etc. (for more on lemmatisation principles, see Cormier et al., 2012). Decisions about which 10 sign tokens to include or exclude from each participant were based on a number of factors. Firstly, only signs that were found listed in the Brien (1992) Dictionary of British Sign Language/English were included. Some signs with a 1 handshape were excluded from this study. These included signs that involved a change from the 1 handshape to some other handshape in citation form (e.g., the compound signs BELIEVE and CHECK) and classifier signs where the handshape can be regarded as an independent morpheme. Like the ASL study, we also ignored thumb-only variants that functioned as non-first person pronominal pointing signs. In these cases, we assumed that the variation in handshape could also be conditioned by the seating arrangement for filming (participants were seated close together and angled such that they were nearly facing each other. In this situation, pointing to the extreme left or right of the signing space towards the addressee was unlikely). We did, however, include the thumb-only variant in first person pronominal points. In addition, we did not include any signs in which the handshape was obscured. The set-up of the filming studio and the three cameras in use during filming afforded us two views of the signer: one camera had both participants in view and each of the remaining two was directed towards one individual in the pair, so we were able to see the hand configuration used in almost all cases. In order to control for frequency of sign tokens, so that a single sign was not over-represented in our final dataset of tokens and to ensure that we had a good mix of individual items, we employed a three strikes rule to token collection. If a single sign occurred three times within one participant s dataset, then all further instances of that sign were ignored. Coding began

9 J. Fenlon et al. / Language & Communication 33 (2013) several minutes into filming so that participants had time to relax in the presence of the cameras. To summarise, after the first three minutes, the first 10 signs to appear that used the 1 handshape in its citation form and fitted our criteria (i.e., it was in the Brien (1992) BSL dictionary, it was not a compound or a classifier sign, and it had not appeared more than three times) were included in the study. The procedure described above produced an initial dataset of 2110 tokens. However, some signs were excluded from the final analysis reported here because we were later unable to determine with certainty whether a given sign had the 1 handshape as its citation form or because we were unable to resolve lemmatisation issues (i.e., whether or not a pair or set of sign variants with similar form and similar meaning should be regarded as separate lexemes). Additionally, as the work conducted here complements concurrent work on a new BSL dictionary, ongoing lemmatisation has revealed evidence suggesting that some signs initially coded as having the 1 handshape in citation form in fact did not (see Cormier et al. (2012) for information on the criteria used to decide a sign s citation form). For example the sign ID glossed as SLOW can be articulated using the 1 or B handshape in its citation form. For these reasons, a total of 17 sign types representing 26 tokens (approximately 1% of the total dataset) were excluded from the analysis reported below leaving 2084 tokens remaining. During coding, each token was coded for the following linguistic factors: handshape category, grammatical category, indexicality, lexical frequency, and handshape of the preceding and following sign. Each was also coded for the social factors described above: gender, age, region, social class, ethnicity and language background. A subset of the pointing signs (i.e., the prononimals and the locatives) were coded for orientation of the hand, and for the orientation of the preceding and following sign. For the category of handshape, each token was placed into one of eight categories represented in Table 2. This handshape table attempts to account for all the possible handshape variants that may be observed in signs specified for the 1 handshape. The handshapes shown in Table 2 do not reflect all of the possible variants in the 1 handshape. As will be seen in the results section, however, the majority of handshape variants are restricted to four categories. The division of categories was partly motivated by the presence or absence of four features: open or closed thumb, open or closed unselected fingers, open or closed index finger, and bent or straight index finger. For the purposes of this study, we judged a sign as being in citation form when the sign was produced with the index finger fully extended, the unselected fingers closed, and the thumb held over or alongside the closed unselected fingers (i.e., the c form shown in Table 2). Please note that, unlike the examples provided in Table 2, not all sign variants deviating from citation form were fully articulated. A form of a sign was categorised as the L variant, for example, when the thumb was no longer touching the unselected fingers, regardless of whether it was fully extended or not. Additionally, the 5 variant did not require all five digits to be fully extended. The index, middle finger and thumb could be extended and the ring and pinky finger closed and the sign would still be classed as 5. Therefore, sign tokens were grouped into the feature category that best represented the token s handshape (according to +/ thumb, +/ unselected fingers, etc.), and the forms shown in Table 2 are simply exemplars of each of these categories. Given the large number of handshapes with varying degrees of finger and thumb extension and the complexity of coding required to represent this accurately, we adapted the simplified coding system used by Bayley et al. (2002) in an attempt to simplify annotation of the data. For grammatical category, signs were coded into the major content grammatical categories of noun (e.g., GIRL), verb (e.g., THINK), adjective (e.g., EASY), adverb (e.g., TOMORROW), depending on their use in sentential context. Function words included pronouns (e.g., PT:PRO1), wh-question signs (e.g., WHY) and other functors (e.g., BUT). As in a previous study (Schembri et al., 2009), we used a number of semantic and morphosyntactic criteria as the basis for our categorisation of sign tokens into these grammatical categories, but in some cases, it was not easy to determine the grammatical function of a specific token. This was more often the case for pointing signs compared to other sign types. Although first person pronouns were unambiguous (as was often the case for second person pronouns), third person pronouns and locative points (i.e., adverbials) proved difficult to distinguish (there is some discussion in the literature about whether strictly pronominal functions for pointing can be distinguished from other uses of pointing in sign languages or indeed in gesture, see Cormier et al., submitted for publication; Evans and Levinson, 2009; Johnston, in press). Third person pronouns were typically identified as a point to the peripheral signing space, away from the conversational partner, and serving as a referent for another person in the discourse. Locative points were identified as a point to a location (e.g., associated with a place-name mentioned in the discourse) or the location at which a topic being discussed was situated. Ambiguous points were often points to a location associated with a referent in a particular location (e.g., person standing in a doorway), making it difficult to determine whether they were primarily locative or pronominal pointing signs. Consequently, these ambiguous points were ignored. Pointing signs functioning as determiners were identified by their syntactic position adjacent to nominal signs and by prosody. If the point (whether it occurred before or after a noun) could be grouped with a noun as a single cohesive prosodic unit, then it was classed as a determiner. For indexicality, sign tokens were grouped into the following two categories: pointing signs and non-pointing signs. Pointing signs were further subdivided into pronominals (first, second, and third person pronouns, both singular and plural forms), as well as locative points (e.g., PT:LOC meaning there or here ), determiners (e.g., PT:DET meaning that or the ), and other uses of pointing (e.g., PT:OTHER which were often temporal points). Thus, the category of pointing signs includes both signs that function as grammatical signs (e.g., pronouns and determiners) as well as signs that may be considered as content signs (e.g., PT:LOC can function as an adverb meaning there ). Initial results indicated that grammatical category and indexicality interacted to a significant degree (i.e., one of the most frequent grammatical categories in the data, pronouns, consisted exclusively of pointing signs) so it was decided to merge the two factor groups into one.

10 78 J. Fenlon et al. / Language & Communication 33 (2013) Lateral Prone Fig. 2. Two categories of orientation used to group pronominal and locative signs. As indicated in previous studies on spoken languages (e.g., Bybee, 2002; Philips, 1984), lexical frequency may be a factor in phonological variation. That is, high frequency words may exhibit greater phonological variation than low frequency words. As such, we coded as high frequency signs in our data all tokens of the top five most frequent items produced with a 1 handshape in a related study of lexical frequency in the BSL Corpus conversational dataset (Cormier et al., 2011): PT:PRO1, PT:PRO3, PT:PRO2, PT:DET and PT:LOC. Together these 5 lexical items represented 1061 tokens, or 51% of our entire dataset. All other tokens (representing the remaining 87 lexical items) were coded together as not high frequency signs. To investigate the effects of the immediate phonological environment, the preceding and following signs were examined and grouped into a category exhibiting the 1 handshape, or exhibiting a different handshape. If no sign immediately followed or preceded the target sign (e.g., when the target sign represented the first or last sign in the participant s turn in the conversation), then we coded this as a pause. We only coded the handshape of preceding/following signs using the same hand as the target sign (regardless of whether there was a shift in dominance). For the orientation study, all non-first person pronoun signs and locative pointing signs were further categorised according to orientation: as lateral, prone (see Fig. 2 above) or other. We also coded whether the orientation of the preceding and following signs matched the orientation of the target sign. The data were coded by the first and third authors, both of whom are deaf native/fluent signers of BSL. Firstly, ten tokens were identified by one researcher and preliminarily coded for all linguistic factors discussed above. The second researcher then examined all the tokens coded in each ELAN file and checked for accuracy and consistency of coding based on the criteria noted above. Any disagreements about coding were discussed, and only tokens in which agreement was reached were included in the analysis. Following completion of coding, 10% (n = 210) of the initial data coded were examined by the second author, a hearing fluent user of BSL, to determine inter-rater reliability. All categories within each token were checked (e.g., handshape, grammatical category, orientation, etc.) and the researcher indicated whether he agreed with the coding assigned or not. This produced an agreement level of 96% (i.e., only 4% of categories being ones that he would have coded differently). Social factors were coded by referring to the demographic questionnaire that each participant filled in prior to taking part. For the statistical analysis, we used the variable rule program Rbrul (Johnson, 2009) which enables us to quantitatively determine the effect of several factors on a binary linguistic variable. In the main analysis, these factors may be linguistic (e.g., grammatical category, the handshape of the preceding sign, etc.) or social (e.g., gender, age) and the binary linguistic variable can be understood as whether the target sign exhibited the 1 handshape or some other handshape (in the other analyses reported here, the linguistic variable might also be a different handshape variant or a specific orientation). Rbrul was designed to improve upon Goldvarb (used extensively in sociolinguistic research, see Tagliamonte, 2006) in that it not only considers fixed effects (such as the linguistic and social factors mentioned above) but random effects as well. In this study, two possible random effects are participant and lexical item. Strictly speaking, tokens are not independent of one other and may be grouped according to these random effects (e.g., 10 tokens can have the same participant in common and, as we see below, 370 tokens have the same sign in common). These random effects themselves may favour or disfavour a variant beyond what can be attributed to fixed effects alone (e.g., a specific lexical item that strongly disfavours handshape variation may appear as a gender effect if the same item is particularly common amongst women in our data). In considering this, Rbrul reports significant results on a given variable only when they are strong enough to discount any possible influence from random effects (something that GoldVarb does not do). A further advantage of using Rbrul is that results are presented in both factor weights, which are used frequently by sociolinguists when displaying results from studies using variable rule programs, and log-odds, which have the benefit of being understood by the wider research community (such as psycholinguistics, psychologists and statisticians). 5. Results 5.1. Distributional results Table 3 provides an overview of the type and frequency of tokens collected for the study together with the number of times each type appeared in citation form.